Mapping Built Infrastructure
Accurate land use land cover (LULC) maps that delineate built infrastructure are useful for numerous applications, from urban planning, humanitarian response, disaster management, to informing decision making for reducing human exposure to natural hazards, such as wildfire.
However, existing products lack sufficient spatial, temporal, and thematic resolution, omitting critical information needed to capture LULC trends accurately over time. Advancements in remote sensing imagery, open-source software and cloud computing offer opportunities to address these challenges.
We demonstrate an open-source approach for mapping built infrastructure in semi-arid systems using remote sensing data and leveraging cloud computing.
Using open source software, including Google Earth Engine (GEE) and QGIS , we:
i) applied detection algorithms to publicly-available remote sensing data and generated annually-resolved mapped products of built infrastructure
ii) derived a time series of change across our region of interest, with which we demonstrated product application by quantifying changes observed and exploring spatial patterns of built infrastructure within our region of interest
iii) spatially and visually compared our time series with publicly available products, including National Land Cover Database (NLCD), Microsoft Building Footprints (MBF), Dynamic World (DW)
Methodology
We created reference data for four separate years by digitizing aerial imagery from the National Agricultural Imagery Program (NAIP), which has a 1-meter spatial resolution. Our reference dataset included six land cover classes: residential, infrastructure, paved surfaces, agriculture, vegetation, and range/scrub.
To develop a training dataset for our Random Forest algorithm, we randomly sampled points from these reference data, generating a pixel dataset used to classify our remote sensing image stacks, contianing multispectral Sentinel-2 imagery, Synthetic Aperture Radar (SAR) Sentinel-1 imagery, ratio-based spectral indices, topographic variables and distance to existing features, such as roads, rivers and railways.
LULC Comparison
We compared our time series with existing LULC products, specifically the National Land Cover Database (NLCD), Microsoft Building Footprints (MBF), Dynamic World (DW).
Our product has a higher thematic resolution than other existing products, and is able to distinguish between built infrastructure types, specifically residental and other infrastructure types.
Access the Mapped Built Infrastructure product and associated materials
Open-source software workflow for mapping built infrastructure in semi-arid systems, including reference data generation, classification, accuracy assessment and validation.
Comparing NAIP aerial imagery to digitized polygons for each LULC class in our workflow ((a) Residential, (b) Infrastructure, (c) Paved, (d) Agriculture, (e) Vegetation, (f) Range/Scrub).
Change over time
Our mapped built infrastructure products are available for the Snake River Plain Level III Ecoregion, in southern Idaho, at a 10 m spatial resolution and an annual temporal resolution for the time series 2015 to 2024.
We show that built infrastructure has increased in extent over time, particularly around Boise, the state capital of Idaho.
Snake River Plain, southern Idaho
Comparing spatial, temporal and thematic resolutions of existing LULC products to our mapped built infrastructure (MBI) products. Both DW and MBI have a spatial resolution of 10 m and thus have an identical pixel size, but these products differ in their thematic resolution. The higher thematic resolution of MBI is illustrated by the three hatched shapes within the black square; however, these do not represent sub-pixel features, but rather serve as a schematic to demonstrate the enhanced thematic detail of our product, compared to the single ‘built’ class in DW. Likewise, the four hatched shapes within the NLCD represent the four ‘developed’ classes.
This work was made possible by the Joint Fire Science Program's Graduate Research Innovation (GRIN) award 22-1-01-06, under the award number L23AC00047 and the Genes by Environment: Modeling, Mechanisms and Mapping (GEM3); a National Science Foundation (NSF) Established Program to Stimulate Competitive Research (EPSCoR) statewide research program in Idaho under the award number OIA-1757324.
I thank the following research assistants for their assistance in generating the reference data: Savannah Canova, Andrew Cesca, Zackery Szymczycha, Karla Rogers, Adrianna Hernandez, Alexandria Serbellon, Micaela Gonzalez, Jacob Scott, Samantha Yonan, Tyler Zafiris and Will Loftin. I thank the three interrelated GEM3 opportunities for college and undergraduate students that financially and academically supported our research assistants: Vertically Integrated Project (VIP), Summer Authentic Research Experience (SARE), and Remote Opportunities for Authentic Research (ROAR). I also thank the broader GEM3 community, the GEM3 Stakeholder Advisory Groups, Peter Olsoy, and Morey Burnham. Last but not least, I acknowledge the Idaho Department of Lands, specifically Tyre Holfeltz, who has supported and provided valuable contributions that shaped this research.
